Elsevier

Journal of Physiology-Paris

Volume 103, Issues 1–2, January–March 2009, Pages 73-87
Journal of Physiology-Paris

Motif distribution, dynamical properties, and computational performance of two data-based cortical microcircuit templates

https://doi.org/10.1016/j.jphysparis.2009.05.006Get rights and content

Abstract

The neocortex is a continuous sheet composed of rather stereotypical local microcircuits that consist of neurons on several laminae with characteristic synaptic connectivity patterns. An understanding of the structure and computational function of these cortical microcircuits may hold the key for understanding the enormous computational power of the neocortex. Two templates for the structure of laminar cortical microcircuits have recently been published by Thomson et al. and Binzegger et al., both resulting from long-lasting experimental studies (but based on different methods).

We analyze and compare in this article the structure of these two microcircuit templates. In particular, we examine the distribution of network motifs, i.e. of subcircuits consisting of a small number of neurons. The distribution of these building blocks has recently emerged as a method for characterizing similarities and differences among complex networks. We show that the two microcircuit templates have quite different distributions of network motifs, although they both have a characteristic small-world property. In order to understand the dynamical and computational properties of these two microcircuit templates, we have generated computer models of them, consisting of Hodgkin–Huxley point neurons with conductance based synapses that have a biologically realistic short-term plasticity. The performance of these two cortical microcircuit models was studied for seven generic computational tasks that require accumulation and merging of information contained in two afferent spike inputs. Although the two models exhibit a different performance for some of these tasks, their average computational performance is very similar. When we changed the connectivity structure of these two microcircuit models in order to see which aspects of it are essential for computational performance, we found that the distribution of degrees of nodes is a common key factor for their computational performance. We also show that their computational performance is correlated with specific statistical properties of the circuit dynamics that is induced by a particular distribution of degrees of nodes.

Introduction

Many complex networks from biochemistry and neurobiology as well as engineering share certain global properties (Newman, 2003, Strogatz, 2001, Watts and Strogatz, 1998), like degree distributions (distribution of the number of edges per node) and small-world properties, i.e. local clustering of edges in a graph while maintaining a short path between nodes. But they often have different local properties, yielding different distributions of stereotypical connectivity patterns for few nodes, called motifs (Milo et al., 2002, Milo et al., 2004, Shen-Orr et al., 2002).

Neurobiological studies have shown that cortical circuits have a distinctive modular and laminar structure, with stereotypical connections between neurons that are repeated throughout many cortical areas (Douglas et al., 1995, Douglas and Martin, 2004, Kalisman et al., 2005, Mountcastle, 1998, Nelson, 2002, Silberberg et al., 2002, White, 1989). It has been conjectured that these stereotypical canonical microcircuits are not merely an artifact of the specific mapping of afferent and efferent cortical pathways or other anatomical constraints like evolutionary processes or development, but are also advantageous for generic computational operations that are carried out throughout the neocortex.

Over the past years detailed statistical data became available that are based on two different experimental methods: dual intracellular recordings in vitro and cell morphology. The first dataset assembled by Thomson et al. (2002) was estimated from 998 paired intracellular recordings with sharp electrodes in slices of somatosensory, motor and visual areas of adult rats and adult cats. It specifies connection probabilities and connection strengths of effectively established synaptic connections between excitatory and inhibitory neocortical neurons, to which we will refer as functional connectivity in this paper. The second dataset assembled by Binzegger et al. (2004) was predicted from bouton and target densities in cat primary visual cortex estimated from three-dimensional cell reconstructions. This dataset does not specify the distribution of functional connections, but rather represents potential synaptic connectivity. The probabilities of synaptic connections between excitatory and inhibitory neurons located in different layers, i.e. layer 2/3, 4 and 5, differ significantly for the functional and the potential microcircuit template (see Thomson and Lamy, 2007). In addition this dataset also includes neurons in layer 6.

We investigate these two cortical microcircuit templates with regard to structural and functional properties. In order to evaluate the computational properties of microcircuit templates we carried out computer simulations of detailed cortical microcircuit models consisting of 560 Hodgkin–Huxley type point neurons and synaptic connections with stereotypical dynamic properties (such as paired pulse depression and paired pulse facilitation) from Markram et al. (1998). Similar to Häusler and Maass (2007), our analysis is based on the assumption that stereotypical cortical microcircuits have some “universal” computational capabilities, and can support quite different computations in different cortical areas. Consequently we concentrate on generic information processing capabilities that are likely to be needed for many concrete computational tasks: to accumulate, hold and fuse information contained in Poisson input spike trains from two different sources (modeling thalamic or cortical feedforward input that arrives primarily in layer 4, and lateral or top-down input that arrives primarily in layer 2/3). In addition we examine the capability of such circuit models to carry out linear and nonlinear computations on time-varying firing rates of these two afferent input streams. In order to avoid rather arbitrary assumptions about the specific type of neuronal encoding of the results of such computations, we analyzed how much information is available about the results of such computations to the generic “neural users”, i.e., to pyramidal neurons in layer 2/3 (which typically project to higher cortical areas) and to pyramidal neurons in layer 5 (which typically project to lower cortical areas or to subcortical structures, but also project for example from V1 back to nonspecific thalamus, i.e. to the intralaminar and midline nuclei that do not receive direct primary sensory input, and through this relay to higher cortical areas, see Callaway (2004)).

In Häusler and Maass (2007) it was shown that the cortical microcircuit model based on the template from Thomson et al. (2002) exhibits specific computational advantages over various types of control circuits that have the same components and the same global statistics of neurons and synaptic connections, but are missing the lamina-specific structure of real cortical microcircuits. Furthermore it was demonstrated that the connectivity graphs defined by this cortical microcircuit template has a small-world property. However we had shown that the degree distribution of neurons is more salient for their computational performance than the small world property.

Here we extend this study by showing that the two cortical microcircuit templates of Thomson et al., 2002, Binzegger et al., 2004 share some global structural properties, like degree distributions and small-world properties, but have significantly different local structural properties, i.e. network motif distributions. A comparison of the information processing capabilities of both microcircuit templates reveals that they have a similar average computational performance but significantly different computational properties for specific tasks. We also address the question which aspect of the microcircuit template of Binzegger et al. (2004) is essential for its computational performance, by scrambling specific aspects of their connectivity pattern in a variety of control circuits. We find that, like for the template of Thomson et al. (2002), the degree distribution of nodes is essential for its computational performance. This result is, besides their similar average computational performance, a second common property of these otherwise quite dissimilar microcircuit templates. We also identify specific properties of the dynamics of the two networks that correlate with their superior computational performance.

Section snippets

Microcircuits and computational tasks

We analyzed cortical microcircuit models based on the laminae-specific connectivity pattern specified by two different cortical microcircuit templates. The first cortical microcircuit template assembled by Thomson et al. (2002) was estimated from paired intracellular recordings with sharp electrodes from 998 pairs of identified neurons from somatosensory, motor and visual areas of adult rats, and visual areas of adult cats. The sampling was made randomly within a lateral spread of 50100μm (

Graph properties of the two microcircuit templates

We analyzed the two data-based cortical microcircuit templates9 shown in Fig. 1, Fig. 2 for their differences and similarities in connectivity structure. In order to evaluate the significance of specific structural features of the two data-based microcircuit templates we compared them with random control circuits which consist of the same number of components, i.e. neurons

Discussion

We found that (1) the microcircuit template by Binzegger et al. (2004) and the microcircuit template by Thomson et al. (2002) have significant small-world properties, but quite different motif distributions; (2) both data-based microcircuits have similar circuit dynamics and average computational capabilities but different computational properties for individual tasks; and (3) the degree distribution is the aspect of the connectivity structure of both data-based microcircuit templates that is

Conclusions

This article has shown that the two available templates for cortical microcircuits have quite interesting structural, dynamical, and computational features. In particular we have shown that it is possible to relate the structure and the (conjectured) computational function of these two microcircuit templates. This positive result will hopefully stimulate further systematic experimental work on the anatomy and physiology of cortical microcircuits, that is needed in order to arrive at a definite

Acknowledgements

We would like to thank Tom Binzegger, Rolf Kötter, Henry Markram, Olaf Sporns, and Alex Thomson for helpful discussions on research related to this article. We also would like to thank two anonymous referees for helpful comments.

References (46)

  • R.J. Douglas et al.

    Recurrent excitation in neocortical circuits

    Science

    (1995)
  • R.J. Douglas et al.

    Neuronal circuits of the neocortex

    Annu. Rev. Neurosci.

    (2004)
  • S. Ganguli et al.

    Memory traces in dynamical systems

    PNAS USA

    (2008)
  • W. Guido et al.

    Relative contributions of burst and tonic responses to the receptive field properties of lateral geniculate neurones in the cat

    J. Neurophysiol.

    (1992)
  • W. Guido et al.

    Response latencies of cells in the cats lateral geniculate neurones are less variable during burst than tonic firing

    Visual Neurosci.

    (1998)
  • A. Gupta et al.

    Organizing principles for a diversity of GABAergic interneurons and synapses in the neocortex

    Science

    (2000)
  • S. Häusler et al.

    A statistical analysis of information processing properties of lamina-specific cortical microcircuit models

    Cerebral Cortex

    (2007)
  • C. Holmgren et al.

    Pyramidal cell communication within local networks in layer 2/3 of rat neocortex

    J. Physiol.

    (2003)
  • M. Kaiser et al.

    Spatial growth of real-world networks

    Phys. Rev. E

    (2004)
  • N. Kalisman et al.

    The neocortical microcircuit as a tabula rasa

    Proc. Natl. Acad. Sci.

    (2005)
  • B.M. Kampa et al.

    Cortical feed-forward networks for binding different streams of sensory information

    Nature Neurosci.

    (2006)
  • R. Kannan et al.

    Simple Markov-chain algorithms for generating bipartite graphs and tournaments

    Random Struct. Algorithms

    (1999)
  • W. Maass et al.

    Real-time computing without stable states: a new framework for neural computation based on perturbations

    Neural Comput.

    (2002)
  • Cited by (35)

    • Relating network connectivity to dynamics: opportunities and challenges for theoretical neuroscience

      2019, Current Opinion in Neurobiology
      Citation Excerpt :

      Furthermore, it has been suggested that only local properties, like the degree distribution, matter [61••,62••]. For example, in [33] it was found that motif structure was not important for certain dynamic properties if one controlled for its effects on the degree distribution. Another approach to studying motifs is to look for specific subfamilies of graphs that have been pre-identified, by theoretical work, as being relevant to neural network dynamics.

    • Population-wide distributions of neural activity during perceptual decision-making

      2013, Progress in Neurobiology
      Citation Excerpt :

      This theory assumes a Gaussian distribution of input currents to each neuron, as generated in a uniformly random coupled network of neurons. An open question is whether the transfer function alone can generate long-tailed firing distributions when the Gaussian assumption is violated, whether by more realistic network topologies that promote localized synchrony (Galan et al., 2008; Haeusler et al., 2009), or by the non-Gaussian statistics of external inputs to the network, as has been demonstrated in rat A1 (DeWeese and Zador, 2006). Side-stepping the assumption of Gaussian inputs, Koulakov et al. (2009) used a simple linear integration and linear transfer function model of an excitatory, recurrent cortical network to study how such long-tailed distributions of spontaneous activity might arise from aspects of network connectivity.

    • Maturation of a Recurrent Excitatory Neocortical Circuit by Experience-Dependent Unsilencing of Newly Formed Dendritic Spines

      2011, Neuron
      Citation Excerpt :

      It will be interesting in the future to directly address whether an excessive NMDAR connectivity is established and pruned back or whether the physical connectivity is matched by experience-dependent functional synaptic and network maturation. Theoretical studies suggest that mature neocortical networks exhibit a small-world network architecture that facilitates information processing by enabling extraction of sensory stimulus from a noisy input (Buonomano and Maass, 2009; Haeusler et al., 2009; Douglas and Martin, 2007; Humphries and Gurney, 2008). Our work suggests that the experience-dependent addition of new functional connections between nearby neighbors on top of an existing low level of connectivity independent of distance, produces small-world architecture.

    View all citing articles on Scopus

    Written under partial support by the Austrian Science Fund FWF # S9102-N13 and project # FP6-015879(FACETS), project # FP7-216593 (SECO), project # FP7-506778 (PASCAL2) as well as project # FP7-231267 (ORGANIC) of the European Union.

    View full text